Traffic congestion detection based on pattern matching and correlation analysis

Single sensor can detect traffic parameters which reflect traffic conditions. However, taking into account the distances between sensors, just relying on data from one single sensor is insufficient to determine traffic congestion. With the development of Remote Transportation Microwave Sensors(RTMS) technology, Chinese highways begin to deploy such devices to monitor traffic flow. However, the distance between two sensors is about 3-4Km, which makes it difficult to process the sparse data from sensors. Using pattern matching and correlation analysis theories, an algorithm for automatic traffic congestion detection (ACD) is presented. Here, one single RTMS is considered as a node. Section average speed and traffic flow are selected as characteristic parameters. According to the traffic congestion reporting information, these two parameters constitute a congestion pattern of this node. So based on large number of previous RTMS data and congestion condition, pattern library of one node could be established. Then, at one single node, Discrete Cosine Transform and Euclidean distance methods were used to match real-time traffic condition and old congestion pattern. When the number of matched node is greater than the preset threhold, the first traffic congestion condition will be found. Next, Pearson correlation coefficient was adopted to measure the correlation between adjacent matched nodes. If the correlation coefficient is much closer to 0, the second traffic congestion condition will be found. Thus, when two conditions meet at the same time, traffic congestion will occur. In order to solve the RTMS data sparse problem, two dimensional linear interpolation methods was introdued to embed virtual nodes into actual nodes, and average speeds of them were calculated respectively. Finally, we selected several section of Beijing-Haerbin highway (G2) for test. 22 RTMS data and traffic event records were collected for 55 days. The experiment results showed that the ACD algorithm based on pattern matching and correlation analysis can effectively and timely detect traffic congestion. With further improved research, these studies will be useful to traffic management when congestion occurs.